Reasoning about Safety of Learning-Enabled Components in Autonomous Cyber-physical Systems.